Author: Keith Ian Price
Date: 17:55:28 04/21/02
Go up one level in this thread
On April 21, 2002 at 20:17:41, Russell Reagan wrote: >On April 21, 2002 at 19:29:39, Will Singleton wrote: > >>All, >> >>[D]5nk1/2R3p1/r4p2/4P3/p2R2N1/7P/1r4P1/7K b - - 0 42 >> >>The question is, my prog wants to play a3, and doesn't see the forced repetition >>until ply 9. I've just thrown away my old buggy rep code, and replaced it with >>a dedicated hash table. The new stuff seems to be debugged and working (maybe), >>but there must be a trick to getting this position earlier. An extension, >>perhaps? >> >>Will > >I assume that the repetition you speak of is after Nxf6 or Nh6, then after pawn >takes knight, Rg5+ followed by Rg7 and perpetual check by Rgg7+ and Rh7+, >correct? > >If so, then it seems like the program should be able to search to ply 7, then it >should be found in the quiescent search. Here's an example: > >1... a3 2. Nxf6+ gxf6 3. Rg4+ Kh8 4. Rgg7 a2 5. Rh7+ This doesn't work because of the knight on f8. Instead it must be 4. Rh4+ Kg8 5. Rg4+, etc. Only Nxf6+ works, due to the knight. > >and then white can draw by repitition. It seems that this wouldn't be caught int >he qsearch originally because of the extra move Rgg7 which is neither a capture >or a check, and if your qsearch only does captures and check extentions, it >won't extend to see the draw by repetition. There is no extra move in the correct variation. > >So the problem here is what to do during a forcing sequence where there are >moves that are not captures or checks, but still forcing, IMO. Perhaps you could >implement an extention when there has been a long series of forcing moves to >allow the program to allow one non-capture or non-check to be extended. In other >words, don't stop qsearch at the first non-capture or non-check, but allow it to >extend a single non-capture or non-check. When you reach the second non-capture >or non-check, you could then terminate the qsearch. > >That's one idea I thought up off the top of my head and I have no idea if it >would cause the search to blow up. It seems like it might cause the search to >blow up at first glance. > >Anyway, the problem seems to be what to do when you have a series of forcing >moves with moves that are at the same time forcing, but also non-captures or >non-checks. In this position, Rgg7 is still a "forcing" move, but it's not so >easy for a computer program to see that I guess. > >Hope this helps, or at least gets some ideas rolling. > >Russell
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.